A factor analysis of long-term bond spreads is performed by decomposing international interest rate spreads into national and global factors. The factors are latent, and are assumed to have GARCH-type specifications as well as exhibiting serial dependence. An indirect estimator is used to compute estimates of the unknown parameters. The sampling performance of this estimator is investigated and compared with an alternative direct estimator based on the Kalman predictor. The factor model is applied to weekly data on long-bond spreads between five countries - Australia, Japan, Germany, Canada and the UK - and the USA over the period 1991 to 1999. The resulting factor decomposition is used to examine the international investor's optimal portfolio decision in a mean-variance framework. Copyright © 2000 John Wiley & Sons, Ltd.